Data assimilation of LAI and biomass into CLM to constrain carbon dynamics in the Arctic and Boreal region

A large amount of carbon is sequestrated in the ­­­Arctic and Boreal (ABoVE) region and carbon dynamics in the area plays a key role in global carbon cycle. Land surface models can be utilized to project future carbon dynamics in the region and the projection will provide a valuable guidance for the government to make corresponding policy to adapt to such change. However, land surface models usually have large uncertainties in predicting carbon dynamics which is attributed to many factors including differences in input forcing, model initial condition, model structure and uncertainties in model parameters. Data assimilation (DA) is an effective way to reduce model uncertainty by providing better model initial condition. The Community Land Model (CLM) is found to highly overestimate leaf area index (LAI) and biomass while underestimate soil organic carbon in the ABoVE region. To correct the bias of the carbon pools, we employ the DART-CLM which is a system that couples data assimilation research testbed with CLM to assimilate high-quality MODIS LAI and a newly-developed biomass observation data into a latest version of CLM and update related state variables of carbon pools in CLM at each assimilation time. We used 40-member ensemble CAM6 reanalysis forcing data to run CLM offline and conducted an assimilation experiment that assimilates LAI and biomass observations into CLM from 2011 to 2019. A corresponding free run experiment (in which DA was not implemented) was conducted for comparison. Compared with the free run, both LAI and biomass as well as the RMSEs of LAI and biomass were reduced in the assimilation run. Meanwhile, DA reduced the overestimation of soil organic carbon in the ABoVE region. We conducted another data assimilation experiment in which only LAI observation was assimilated into CLM. To investigate the impact of initial condition on forecasting carbon pools, we did two forecast experiments that used the two different initial conditions achieved in the two assimilation experiments. It turns out the forecast run starting with the initial condition from the experiment assimilating both LAI and biomass has a longer persistence of accurate model forecast of both short-term and long-term carbon pools. Our DA work provides promising initial conditions to get reliable projections of carbon dynamics in ABoVE region.